…typed-vio shape

Tests were left on the untyped shape by the initial L2 landing — go vet failed
because:
- AnyVersionedWrite is an interface (Header(), ValAny()), tests used direct
  w.Address / w.Path / w.Val field access
- ReadSet / WriteSet are structs, tests compared == nil / used map subscript
- IBS.SetNonce / SetCode and stateObject.SetNonce / SetCode now take a
  tracing.NonceChangeReason / CodeChangeReason — tests called with 2 args
- chain.Config.ChainID is *uint256.Int on main; chain-end test used
  big.NewInt(1)
- SD.Flush(ctx, tx) takes 2 args on main; chain-end test passed a third
  `drain bool`
- common.Address.Value() is array-typed, so direct slicing isn't addressable
  in a composite call — bound to a local var first

Files refreshed from chain-end (db9705dd16) and adapted to main's signatures
with a paren-balanced auto-insert of tracing.NonceChangeUnspecified /
tracing.CodeChangeUnspecified at every 2-arg call site.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…imeline log

Adds env-gated [sched_timeline] log emission so the per-block
concurrency-over-time curve can be reconstructed offline. Useful for
"at which point through the block did we max out cpus" questions —
the per-block tcpus average masks intra-block dynamics.

Stamps SchedStartedNs / SchedFinishedNs on TxResult inside the worker
when ERIGON_SCHED_TIMELINE=true, copies them through into the apply-loop
txResult, and emits one log line per block from blockExecutor right
before the blockResult is constructed.

Cost when disabled: zero (Guard on dbg.SchedTimeline).
Cost when enabled: 2 × time.Now().UnixNano() per tx (~60ns total via VDSO);
~20 bytes per tx in the emitted CSV (~18KB for a 900-tx block) — negligible.

Format:
  [sched_timeline] blk=N base_ns=B cnt=K data="idx:start_delta_ns:dur_ns,..."

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…tion/commitment: tighten vio-shim comments

Drop multi-paragraph shim comments per CLAUDE.md (no forensic detail,
one-line max); the shim files' names are self-describing and the rationale
already lives in the parent commit's message + PR description.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…dResult

blockExecutor.blockIO.Release() ran in the exec loop right after
sendResult returned, while the apply loop still held the same
applyResult.TxIO reference. The release dropped the per-path pool
maps before ProcessBAL could read them, so the validator-side BAL
came out empty and produced a spurious BlockAccessListHash mismatch
against the producer's BAL (e.g. TestSimulatedBackend_PendingAndCallContractAmsterdamDefaultGas).

Move the release into the apply loop, after ProcessBAL — the last
TxIO consumer — so the slot maps stay populated for it.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Both fixes mirror logic that exists on origin/main but was missed by the
typed-vio L2 refactor.

1. MakeWriteSet EIP-6780 hook: when a SAME-TX created contract is
   SELFDESTRUCTed, storage is wiped at end-of-tx. The BAL must report
   the dirty slots as reads, not changes. Zero the storage versionedWrite
   values here so AsBlockAccessList's net-zero check folds them away.

2. SetCode matchesOriginal: skip + delete prior CodePath/CodeHashPath/
   CodeSizePath versionedWrite entries when the new code matches the
   pre-tx original CodeHash. Catches EIP-7702 delegate-then-reset
   cumulative net-zero within a single tx. Guarded against newlyCreated
   stateObjects whose .original is the pre-creation snapshot.

Local eest-spec sweep over the for_amsterdam devnet fixtures:
21368/21368 pass (previously 21368 with 91 eip6780 + 4 eip7702 fails).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

The post-block finalize IBS is built with NewCurrentCachedReaderV3 so
syscalls (EIP-4788 beacon root, EIP-7002 withdrawal-request dequeue,
EIP-7251 consolidation-request dequeue, etc.) see the latest per-tx
writes accumulated in BlockStateCache. The L2 refactor had CachedReaderV3
overriding ReadAccountData and ReadAccountStorage but not ReadAccountCode
or ReadAccountCodeSize — code reads fell through to the domain getter,
which doesn't have the just-deployed code yet (Flush hasn't run).

When block 1 of a fork transition deploys a system contract via CREATE,
the post-block syscall's GetCodeSize reads through the cache miss and
sees an empty code, fails "[EIP-7002] Syscall failure: Empty Code",
and the block can't finalize:

    apply loop exited but 1 block(s) had tx-results without a blockResult: [1]

Add GetCurrentCode on BlockStateCache and ReadAccountCode /
ReadAccountCodeSize on CachedReaderV3, matching origin/main's shape.

Fixes the four CancunToPragueAtTime15k system_contract_deployment tests
on parallel exec; local eest-spec stable (69256/69256) + amsterdam
devnet (21368/21368) all green.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…ways false

exec3_parallel set InMemHistoryReads=true on entry and unconditionally
restored it to false on exit. That broke any post-exec caller that
expected history reads on the shared mem batch: the engine-x runner's
restore-state path calls GetAsOf and fails with

    GetAsOf called on TemporalMemBatch with inMemHistoryReads disabled

Capture the prior flag value and restore it on exit, matching origin/main's
shape.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

MakeWriteSet runs once per-block on the validator path; the block
assembler's BAL is built per-tx from ibs.TxIO() and never goes through
MakeWriteSet. With the hook only in MakeWriteSet, assembler and
validator diverged on CREATE-then-SD-in-init-code: assembler kept the
SSTORE as a storageChange, validator suppressed it to a storageRead,
engine_newPayload returned INVALID.

Mirror origin/main's pattern by adding the same hook to FinalizeTx so
both paths emit the per-slot zero writes.

Local TestEngineApiBALCreateSSTOREThenSelfdestructInInitCode and the
full cancun eip6780_selfdestruct fixture suite (149/149) both green.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

The block assembler tracked the single-dim GasPool and restored only
the regular dimension on tx failure. Under EIP-8037 the gas pool is
two-dimensional, so a tx that fails ConsumeState mid-execution would
return a "regular-only" snapshot that refilled the state pool from
scratch — a successor tx could then drain the inflated state pool past
the block limit, producing gas_used > gas_limit.

Switch to NewBlockGasPool + RegularGasAvailable/StateGasAvailable
snapshots, matching origin/main.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

… not cell pointer

ReadStorageCell returned a *WriteCell[uint256.Int] for the caller to read
cell.Value AFTER releasing the read lock. flushVWLocked, called under the
write lock, mutates that same cell.Value when re-flushing at the same
(txIdx, incarnation) pair. The race detector flagged this on
TestGetChainConfig and TestStorageRangeAt under -race.

Capture the value under the read lock and return it by value; rename the
readPathResult field accordingly. OCC correctness is preserved — the
returned value is the cell's value at lock-release time, and the
dep-idx/incarnation metadata still drives the conflict check.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Test #119 'Re-Org Back into Canonical Chain ... Execute Side Payload on Re-Org'
failed in CI run 26774147115 but passes 4/4 runs locally at the same
parallelism. Suspected CI-environment timing flake.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Replace the changeset-accumulator dump (cs is nil during fork validation,
which is exactly the path hive Re-Org Back into Canonical Chain hits)
with a targeted dump of ZeroAddr (the hive test's hot recipient for 50
txs/block) and the block coinbase. Reading via Domains().GetLatest goes
through sd.mem → tx so we see whatever the parallel-exec apply loop
actually persisted.

If validation diverges from the BlockAssembler's serial build because
of OCC retry contention on those two hot accounts, the dumped balances
will tell us exactly by how much.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Replace the raw bytes dump with a decoded balance + nonce so we can
read the divergence directly. ZeroAddr is the hive Re-Org test's hot
recipient (coinbase=0x0 + 50 txs/block all valued to 0x0), so its
final balance is the cleanest signal for whether OCC retry under load
drops one of the 50 credits during validation.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

PR #21536 hit a fully-green CI on cd9078e (87 pass, 0 fail). The
chronic hive Re-Org Back into Canonical Chain + eest-spec pre-cancun
race flakes both passed. Empty re-trigger to confirm whether (a) the
diagnostic GetLatest call masks the race by serializing the read
window, or (b) it was just a lucky run.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Sync.Run iterates stages by s.currentStage++ starting from whatever value
currentStage holds on entry. The exit-time defer resets it to 0, which is
correct for a single-threaded caller — but pe.sync is shared between the
forward Run path and the background RunPrune goroutine spawned from
updateForkChoice when --fcu.background.prune is on. RunPrune iterates by
calling SetCurrentStage(stage.ID) inside pruneStage to update the log
prefix. When a forward FCU arrives while the previous FCU's BG prune is
still iterating, the forward Sync.Run picks up at whatever stage position
the prune was at and skips the earlier stages (Snapshots/BlockHashes/
Senders). Execution then no-ops because its prevStageProgress (Senders)
is behind the FCU's target, the head fails to advance, and the
"bad forkchoice" warning fires at the end of updateForkChoice.

Surfaced as the hive "Re-Org Back into Canonical Chain, Depth=5" (test=9)
flake locally repro'd at ~5% under 100-iter stress at parallelism 12.
Same fix also closes the "Re-Org Back into Canonical Chain, Depth=10,
Execute Side Payload on Re-Org" variant which had previously surfaced as
intermittent "Wrong trie root of block N" in CI — same race, different
symptom depending on whether a NewPayload sits between the back-FCU and
the forward-FCU.

Two soaks at 100/100 clean post-fix.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…o typed-vio

origin/main moved 21 commits forward since the last sync; #21590 (parallel-exec
SD-revival and metamorphic-CREATE2 fixes, da0a479) touches the same files
the typed-vio L2 refactor does and produced 5 conflicts. Resolutions:

- execution/state/intra_block_state.go::CreateAccount — drop the
  `(source, version)` inheritance for synthetic IncarnationPath/BalancePath
  reads (each defaults to (StorageRead, UnknownVersion), only promoted to
  MapRead when the path's own versionMap cell exists). Gate the
  recordWriteIncarnation on contractCreation=true. Use typed ReadIncarnation /
  ReadBalance helpers instead of generic Read.
- execution/state/intra_block_state.go::accountRead — demote a sub-field
  MapRead promotion to (StorageRead, UnknownVersion) when AddressPath has no
  cell, via typed ReadAddress.
- execution/state/intra_block_state.go::getVersionedAccount — already had
  the SD-revival check (>=destructTxIndex for AddressPath, > for sub-fields).
- execution/state/rw_v3.go::LightCollector.UpdateAccountData — emit
  IncarnationPath only on up-revs (`>` not `!=`) so a value-transfer to a
  same-block SD'd address can't clobber the SD-side cell. Retained typed
  &VersionedWrite[uint64]{...} form.
- execution/state/versionedio.go::ReadAccountData — same-tx metamorphic
  recreation check (AddressPath at >=destructTxIndex first, sub-field >
  fallbacks) using revivalLimit = txIndex - 1.
- execution/state/versionedio.go::addStorageUpdate — added the EIP-7928
  no-op filter (skip when value equals slot's last recorded write); kept
  the free-function signature.
- execution/state/versionmap.go — added AnyDoneBoolWriteEquals (scoped to
  SelfDestructPath given the typed-vio per-path WriteCell[T] layout).
- execution/stagedsync/exec3_parallel.go::normalizeWriteSet — added the
  SD-then-revival history-scan branch (latest SelfDestructPath may be false
  after revival; AnyDoneBoolWriteEquals catches the prior true).
- execution/state/versionedio_test.go — ported the 4 #21590 regression
  tests (TestAccountRead_BalancePathPromotion / TestCreateAccount_
  SyntheticIncarnationStamp / TestGetVersionedAccount_PriorTxSelfDestruct /
  TestGetVersionedAccount_SameTxMetamorphicRecreate) and the 2 storage
  no-op/RM-DEP-fallthrough tests, re-typed to VersionedWrite[T]/
  VersionedRead[T] and per-path Set helpers.

Not ported: #21590's calcFees coinbase-emit fix (EIP-161 empty-removal of
zero-balance coinbase). vio-on-main's calcFees uses the typed CollectorWrites
machinery and a different flow; the fix would need a separate semantic merge.
This affects coinbase emptiness on EIP-161 forks but does not affect the
SD/CREATE2 reincarnation class the test_recreate eest-race flake exercises.
Tracked as a follow-up.

Also addresses sudeepdino008's review comment: removed the SetMaxCollationTxNum
shim in db/state/aggregator_vio_shim.go and its one call site in
execution/stagedsync/exec3.go (the upstream method no longer exists on main).

Verification:
- make erigon integration clean
- make lint clean (2 runs)
- make test-short clean (including the 4 ported regression tests)